combining patronage and merit in public sector recruitment

Combining Patronage and Meritin Public Sector Recruitment

Sarah Brierley, London School of Economics and Political Science

When do politicians in developing democracies prioritize meritocratic recruitment over patronage hiring for public sector

jobs? I distinguish between low- and high-skilled positions and argue that the former are valuable for sustaining party

machines, while manipulating the latter can undermine state performance. Accordingly, politicians will interfere to ensure

their copartisans are hired to low-skilled jobs but select bureaucrats for high-skilled positions based on meritocratic

criteria. I test my argument using novel bureaucrat-level data from Ghana (n p 17; 942) and leverage a change in the

governing party to investigate hiring patterns. The results suggest that partisan bias is confined to low-ranked jobs. The

findings shed light on the mixed effects of electoral competition on patronage identified in prior studies: competition may

dissuade politicians from interfering in recruitment to high-ranked positions but encourage them to hire partisans for low-

ranked positions.

Whether civil servants are hired based on merit orpolitical criteria has broad implications for statecapacity and the overall health of democracy (Ged-

des 1994; Grzymala-Busse 2007; O’Dwyer 2006). Patronagerecruits are likely to be less competent and also not essentialto the running of the state. This type of hiring can place anunnecessary strain on the public purse, and undermine in-vestment in other vital areas, such as capital infrastructure. Itcan also perpetuate a broader clientelistic political economy.Once hired, copartisan bureaucrats can help politicians allo-cate scarce public resources in suboptimal ways.1 Partisan ap-pointees may also worsen the provision of public services(Colonnelli, Prem, and Teso 2017) and increase levels of cor-ruption (Oliveros and Schuster 2018).

Given these high stakes, it is important to understandwhen politicians in developing democracies are more likelyto rely on meritocratic versus partisan recruitment. Prior re-search has not been able to adequately address this questionbecause, by using changes in legislation to signal the onset ofmeritocracy (Geddes 1994), it treats the concept of meritoc-racy as dichotomous. Other research similarly assumes thequestion away by portraying postcolonial states as neopatri-monial, a situation in which all public sector jobs are often

Sarah Brierley ([email protected]) is an assistant professor in government atWC2A 2AE, United Kingdom.

Data and supporting materials necessary to reproduce the numerical res.harvard.edu/dataverse/jop). An online appendix with supplementary material is

1. See Golden and Min (2013) for a review of the literature on distributive

The Journal of Politics, volume 83, number 1. Published online July 10, 2020. hq 2020 by the Southern Political Science Association. All rights reserved. 0022-

assumed to be distributed politically (Bayart 1993; Chabal andDaloz 1999; Van de Walle 2001).

Despite this dominant narrative, in practice most coun-tries lie on a continuum between complete meritocracy andwholesale partisan interference (Grindle 2012). I build onthis insight and consider when the benefits of patronagehiring are outweighed by the costs. I theorize that the costs ofnoncompetitive recruitment vary across public sector posi-tions, and therefore that politicians may support merito-cratic recruitment for some posts and actively interfere inhiring for others. Meddling in hiring for professional posi-tions, such as budget analysts, engineers, and planning offi-cers, is costly to politicians. It can hurt incumbents elector-ally because it can impact state performance. In contrast,interference in low-ranking public sector jobs, such as san-itation officers, laborers, and security guards, poses a lowerrisk to state performance. Furthermore, politicians can awardlow-ranked positions to party brokers in order to sustain partymachines. In many developing countries party brokers lackthe necessary educational background to succeed in high-ranked positions. In short, incumbent politicians in develop-ing democracies can pursue a dual strategy of hiring bureau-crats for high-skilled jobs based on meritocratic criteria and

the London School of Economics and Political Science, Holborn, London

ults in the article are available in the JOP Dataverse (https://dataverseavailable at https://doi.org/10.1086/708240.

politics and partisan bias in the allocation of public resources.

ttps://doi.org/10.1086/7082403816/2021/8301-00XX$10.00 000

000 / Combining Patronage and Merit in Public Sector Recruitment Sarah Brierley

bureaucrats for low-skilled positions according to partisancriteria.

To test my argument, I assembled unique data on theuniverse of over 40,000 bureaucrats working in 200 localgovernments in Ghana, a stable democracy in West Africa.Such fine-grained data from a developing country are oftennot available or are very hard to obtain, which is why only ahandful of studies has been able to use similar microlevel data.2

Electoral politics in Ghana are dominated by the NationalDemocratic Congress (NDC) and the New Patriotic Party(NPP). Since the restoration of democracy, the country’s na-tional elections have become increasingly competitive. TheNDC won the 2008 presidential election—the focus of thisarticle—by less than half a percent.3 In the analysis, I exploitthis change in the ruling party to investigate hiring patterns.

The data on bureaucrats include information on the hiringdate of each employee. I use the first term (2005–8) as abaseline for comparison with hiring patterns in the secondterm (2009–12). My study sample consists of all bureaucratshired during these two terms (n p 17;942). If the conven-tional wisdom of blanket patronage hiring is correct, wewould expect the new ruling party to favor its copartisans forall positions, at the expense of opponents. The data includeinformation on bureaucrats’ individual characteristics. In theanalysis, I use bureaucrats’ ethnicity and home region asproxies for their likely partisanship.4

Disaggregating the data between high-skilled and low-skilled positions, I find no evidence of partisan hiring forhigh-skilled positions. High-skilled recruits also becomebetter qualified over time, which suggests a preference forcompetence. Conversely, a change in ruling party is associ-ated with a 10 percentage point increase in the probabilitythat a copartisan will be hired to a low-skilled post. This in-crease is equivalent to about 635 extra public sector jobs be-ing awarded to government copartisans, at the cost of overUS$3 million over a four-year term.5 Because the data onlycontain information on local bureaucrats, this bias likelyrepresents a fraction of the aggregate number of partisanhires during the study period. While the analysis does not

2. The empirical study of bureaucratic recruitment in developingcountries is a burgeoning field in political science and economics. Relevantpapers include Colonnelli et al. (2017), Hassan (2016), Iyer and Mani(2012), Pierskalla and Sacks (2019), Sigman (2015), and Xu (2018).

3. The NDC won with 50.2% of votes, compared to 49.8% for NPP.4. In the section on coding civil servants’ partisanship, I discuss how I

infer partisanship from these two variables.5. The total number of bureaucrats hired to low-skilled positions

between 2009 and 2012 was 6,359. See table C.1. To calculate the cost of635 bureaucrats, I assume a (conservatively low) monthly salary of US$100.

confirm that such bias led to the appointment of bureau-crats who were not qualified to take low-ranked posts, itdoes suggest that bureaucrats’ identity influenced who thenew governing party hired.

My theory suggests that politicians reward party brokersfor their support by securing low-skilled jobs for them. Icombine quantitative and qualitative data to substantiatethis claim. The qualitative evidence suggests that grassrootsbrokers in Ghana are motivated to work for parties in returnfor material benefits, including public sector jobs. The dataalso provide evidence of strategic hiring: politicians allocatemore new jobs to districts where brokers attract new voters.

This article makes three contributions. First, the resultscontribute to the literature on public sector development. Itis one of the first empirical studies in political science to useadministrative data to document trends in public sectorhiring practices in a developing country (see also Hassan2016; Pierskalla and Sacks 2019). My theoretical approachalso sheds light on what appear to be inconsistent findingsregarding the effect of electoral competition on patronagehiring. While many scholars assert that competition can pro-mote meritocracy (Geddes 1994; Ting et al. 2013), others findthat competition can encourage clientelism, and swell theranks of the public sector (Driscoll 2017; Lindberg 2003; Pier-skalla and Sacks 2019). My theory suggests that electoralcompetition may have both of these effects: it may dissuadepoliticians from interfering in hiring for high-ranked posi-tions, while encouraging them to recruit partisans to low-ranked positions.

Second, I advance the literature on measuring meritoc-racy. Past studies have primarily used either legislation (Ged-des 1994) or surveys of experts or bureaucrats (Kopeckỳ 2011;Sigman 2015) to measure this concept. My approach —whichcombines bureaucrat-level data with a change in the rulingparty—is more objective than survey-based methods. Fur-thermore, the data allow us to analyze de facto practices ratherthan de jure protocols.

Third, I contribute to the burgeoning empirical literatureon party brokers (Calvo and Murillo 2013; Larreguy, Mar-shall, and Querubin 2016; Stokes et al. 2013). My work com-plements prior studies that show brokers often hold or seekto obtain public sector positions (Oliveros 2016) and advancesthis literature by considering the types of jobs that brokersare likely to hold.

THEORY OF DUAL HIRING: COMBINING MERITAND PATRONAGEGovernments in low-income countries dedicate about athird of public budgets to bureaucrats’ salaries (Clements

9. Similarly, Gottlieb et al. (2018) argue that incumbent parties createnew administrative units to coopt strong community brokers in areas

Volume 83 Number 1 January 2021 / 000

et al. 2010).6 In African countries, the wage bill is often thegovernment’s single biggest expense: it represents 55% ofpublic expenditure in Kenya, 45% in Ghana, and over 33% inSouth Africa.7 While the seminal literature on the Africanstate suggests that governments distribute most public sectorjobs on the basis of political criteria (Bayart 1993; Chabal andDaloz 1999), these claims have not been tested empiricallysince the reintroduction of multiparty elections in the early1990s. In this article, I develop a theory of when politicians indeveloping democracies prioritize meritocratic recruitmentover patronage hiring for public sector jobs.

Benefits to politicians of distributing low-skilledpublic sector positions to partisansI propose that the costs and benefits of patronage hiring topoliticians in developing democracies vary according to thetype of public sector job. Certain types of jobs are especiallyvaluable to politicians. In particular, politicians who operatein clientelistic polities will value public sector jobs that theycan award to grassroots party brokers. Awarding jobs tobrokers has a multiplier effect on votes: the politician gainsthe support of the broker and her clients. US party leadersused jobs in the postal service and customs houses to buildand sustain party machines during the nineteenth century(Carpenter 2001). For example, in Chicago, political partiesdistributed jobs to precinct captains who worked to get outthe vote and dispense private goods to citizens (Wilson 1961).8

Historical accounts of the US transition toward meri-tocracy suggest that it took decades for low-ranking jobsoutside the capital to be taken out of the hands of politicians.Patronage was sustained for most field positions and topadministrative jobs in the capital (Skowronek 1982, 69). Postoffices employing over 50 persons were subject to the Pen-dleton Act of 1883. However, in 1896 there were still 76,000fourth-class postmaster positions that were not covered bythe act and were therefore available for parties to distributeto loyalists (Skowronek 1982, 72). In contrast, clerical jobsbased in the capital were easy for politicians to give up.

The practice of awarding public sector jobs to those whocan mobilize voters on behalf of the incumbent party hasbeen documented in other contexts. O’Dwyer (2006) arguesthat in new Eastern European democracies, citizens were de-mobilized after years of communist rule such that mass parties

6. See table A.1.7. On Kenya, see Aljazeera (2014); on Ghana, see Africa Confidential

(2018); and on South Africa, see BusinessTech (2018).8. In the United States, another major benefit of patronage hiring was

the contributions (or assessments) that office holders gave to the party,which constituted 2%–10% of their salaries (Johnson and Libecap 1994,15).

with fee-paying memberships are not possible. He maintainsthat incumbent politicians instead give public sector jobs toparty activists (529). Callen, Gulzar, and Rezaee (2020) arguethat politicians in Pakistan provide patronage to doctors be-cause they act as important political mediators in rural areas atelection time. An original survey of party brokers in Argentinashows that 30% of brokers have public sector jobs (Stokes et al.2013, 99). In Africa, scholars argue that politicians create newlocal-level units precisely so they can distribute new admin-istrative and political positions to local brokers (Hassan andSheely 2017) and party supporters (Green 2010).9

While party brokers have significant leadership skills andvaluable ties to local elites (Brierley and Nathan 2020), indeveloping countries they are often not particularly welleducated. Indeed, many intermediaries work for parties forthe material benefits they can extract to supplement theirincomes (Bob-Milliar 2012). A survey of grassroots partybrokers in Ghana shows that they are “characteristically veryyouthful but poorly educated or without formal education”(Bob-Milliar 2012, 670). Thus, politicians can usually onlyreward them with low-skilled positions.10 Because middle-classcitizens with university degrees have fewer financial incentivesto work for parties, they are less likely to become brokers. Insummary, politicians in developing democracies who want tosustain party machines will seek to exercise control over hiringfor low-skilled jobs in the public sector, which they can use toreward grassroots party brokers.

Costs to politicians of distributing high-skilledpublic sector positions to partisansWhile the benefits of patronage appointments include theability to extract loyalty and campaign work from public of-ficials, patronage hiring also has at least two important costs.First, it can lead to the inefficient functioning of the state.Second, as the size of the public sector grows, interfering inrecruitment processes becomes increasingly time intensive forpoliticians. These costs to politicians are likely to be greater forhigh-skilled jobs than for low-skilled jobs, which can in-centivize politicians to recruit professional bureaucrats on the

where the party is not already strong. While they do not suggest that localgovernment jobs are distributed directly to brokers, they imply thatthe incumbent party uses local government units to mobilize votes viacommunity-level brokers.

10. The jobs that party brokers can reasonably perform will vary ac-cording to the country’s level of development. In richer countries, wherethe average level of education will be higher than in poorer countries,brokers may be educated enough to hold mid-ranked positions in thebureaucracy.


basis of meritocratic criteria.11 In developing democracies,politicians often retain discretionary control over bureaucrats’careers, which they can leverage to influence their work. Suchdiscretion allows politicians to get the best of both worlds:high levels of competence and loyalty.

Regarding state performance, recruiting bureaucrats tohigh-skilled jobs on the basis of partisan, as opposed to mer-itocratic, criteria can lead to hiring less competent bureaucrats.Partisan recruits are likely to underperform their duties andto be more corrupt (Oliveros and Schuster 2018). Indeed, mer-itocratic recruitment has been identified as the most impor-tant factor in improving bureaucratic performance (Rauchand Evans 2000). Bureaucratic underperformance can lead tomacroeconomic instability and the inefficient delivery ofpublic services. Politicians are likely to be concerned with poorstate performance if voters consider government performancewhen voting.

Evidence suggests that even in relatively clientelisticpolities, some voters consider economic management andpublic service delivery when they vote. The incumbent party’sability to deliver on macroeconomic policies has been shownto be an important determinant of vote choice in multipleregions, including in Africa (Bratton, Bhavnani, and Chen2012), Latin America (Lewis-Beck and Ratto 2013; Robertsand Wibbels 1999), and Eastern Europe (Roberts 2008). Per-formance voting has also been demonstrated in individualAfrican countries such as Zambia (Posner and Simon 2002),Ghana (Hoffman and Long 2013; Lindberg and Morrison2008; Youde 2005) and South Africa (Ferree 2006; Mattes andPiombo 2001).12 Regarding development outcomes, votersprefer politicians who deliver local public infrastructure such asroads, schools, and health clinics (Carlson 2015; Harding 2015;Ofosu 2019; Weghorst and Lindberg 2013).

Electoral pressures on politicians to deliver development,combined with the difficulty of instituting wholesale merit-ocratic recruitment practices, can lead politicians to establishbureaucratic “pockets of efficiency” (Geddes 1994). Thesenational-level agencies operate outside the federal bureau-cracy and recruit personnel on the basis of merit (61). Ap-

11. My argument does not apply to top positions in the public sector suchas heads of executive agencies and ambassadors, which remain patronageappointments in most countries (Fisman and Golden 2017, 39). Unlike otherpublic sector positions, these top-tier jobs are usually not subject to tenure.

12. For example, Posner and Simon (2002) find that “regardless ofbackground, respondents who expressed dissatisfaction with the economywere 10–15 percentage points less likely to vote for [President] Chilubathan their more satisfied counterparts” (319). Hoffman and Long (2013)conclude in Ghana that “demographic and ethnic factors are far lessimportant than respondents’ beliefs about the parties, candidates, theNPP’s performance, and economic conditions” (139).

plying this argument to the local level—the empirical focus ofthis article—politicians may also have an incentive to recruitcompetent bureaucrats to work in local governments to effi-ciently deliver local public goods and services to voters. Giventhe smaller size of the workforce in local governments, themarginal value of a highly trained bureaucrat is higher in localgovernment offices compared to the national bureaucracy.

Politicians’ concerns about the potential electoral reward(or punishment) of better state performance are likely to beintensified in competitive—or increasingly competitive—democratic environments. The positive relationship betweenelectoral competition and the adoption of meritocratic re-forms has been demonstrated both theoretically (Geddes 1991;Ting et al. 2013) and empirically (Geddes 1994; Grindle 2012;Grzymala-Busse 2007; Ting et al. 2013).13 Geddes (1991)argues that when two political parties are roughly equallypopular, it becomes advantageous for the incumbent to adoptreforms in order to capture the electoral benefits of institutingmeritocracy. Grzymala-Busse (2007) argues that robust com-petition leads to monitoring by opponents. As a result, gov-erning parties avoid exploitative state practices to preemptivelyprotect against opposition criticism. In short, electoral incen-tives can encourage merit-based hiring due to the resultinggains (or losses) in incumbent vote share.

A second cost to politicians of interfering in high-skilledpublic sector positions is the time it takes to distribute andmonitor these appointments. Urbanization and industriali-zation lead to an increase in the size of the public sector, andbureaucratic tasks become increasingly complex. As the publicsector gets larger it becomes more costly for politicians tomanage recruitment processes. As the number of positionsincreases, politicians are required to spend more time ensur-ing that these positions are filled by loyalists. At some point,the time politicians spend interfering in hiring processesoutweighs the benefits. Scholars of the United States haveargued that the growth in the size of the federal labor force was“the overriding factor that changed the way in which federalpoliticians viewed patronage” (Johnson and Libecap 1994, 13).

The costs of patronage hiring on politicians’ time are likelyto be greater for high-skilled positions than for low-skilledjobs. This is because politicians can delegate the recruitment

13. In contrast to this literature, one can reasonably argue that elec-toral competition can lead incumbents to prefer an inefficient bureaucracyto an efficient one because, should the incumbents lose office, their op-ponent can use a highly efficient bureaucracy against them in the future.However, incumbents must weigh this potential should they lose, to theincentive they have to take steps to remain in office. In situations wherethe likelihood of remaining in office is strengthened by performing well,politicians are likely to prioritize strengthening the state.

14. The NDC won the 2012 election by a margin of 3%.15. High levels of electoral competition may be present in countries

with both stable and more fluid party systems.


of personnel for low-skilled positions to local party organi-zations, which can identify grassroots brokers who are suit-able for these roles. It may be harder for local party elites toidentify relevant people for high-skilled positions. Instead,politicians can delegate the recruitment of bureaucrats forhigh-skilled positions to other bureaucrats, who will then or-ganize competitive examinations, develop interview proce-dures, and establish minimum qualification standards foreach job. After bureaucrats have developed rules for merito-cratic hiring it becomes more difficult for politicians to inter-vene. Such interventions are likely to be visible, which cangenerate criticism from other civil servants, political oppo-nents or from nongovernmental organizations, and result innegative media coverage.

Finally, there is an overlooked reason why politicians mayhave an incentive to prioritize professional competence overpartisan loyalty: politicians often continue to be able to in-fluence the work of bureaucrats even when they do not hiretheir copartisans. Empirical research from India and Ghanasuggests that politicians may be willing to allow competitiverecruitment to high-skilled positions because they know theycan still interfere in bureaucrats’ careers to influence theirbehavior (Brierley 2020; Iyer and Mani 2012). Despite highlycompetitive examinations to enter the top ranks of the civilservice in India, for example, politicians continue to influ-ence the actions of civil servants through their control overgeographic transfers (Iyer and Mani 2012; Wade 1982). Pol-iticians can use transfers, or the threat of transfers, to influenceoutcomes such as where new local public goods are placed.This suggests that politicians can recruit highly educatedbureaucrats to high-skilled positions without stymieing theirattempts to politicize the distribution of state resources.

In summary, performance voting incentivizes politiciansto develop a preference for hiring competent professionals.This preference will intensify as electoral competition in-creases because small electoral gains or losses become po-litically consequential. Pressure on politicians to delegate bu-reaucratic recruitment to other bureaucrats will also occuras states become larger and more complex. Because politi-cians can delegate the recruitment of bureaucrats for low-skilled jobs to party machines, pressures on politicians willbe more significant for high-skilled positions. Finally, poli-ticians are likely to feel more comfortable delegating recruit-ment to high-skilled positions to other bureaucrats becausethey have additional tools—besides hiring—that they can le-verage to control the work of bureaucrats once appointed. Thisarticle tests the hypothesis that politicians in clientelistic de-mocracies are more likely to hire public employees on the basisof partisan loyalty for low-skilled jobs rather than for high-skilled positions.

GHANA’S ELECTORAL ENVIRONMENT AND SCOPECONDITIONS OF THE ARGUMENTI evaluate my argument using a novel data set of bureaucratswho work in local governments in Ghana. Ghana’s electoralsystem has supported three peaceful transitions of executivepower (in 2000, 2008, and 2016). Two political partiesdominate the electoral landscape, the NPP and NDC. Na-tional elections are competitive and have become increas-ingly so since democracy was reinstituted in 1992. I focus onthe 2008 election, which the NDC won by a margin of lessthan 0.5% of the votes. This election followed two priorelections that saw a significant decrease in the margin ofvictory of the presidential victor; while in the 2000 electionthe NPP presidential candidate won by a margin of 14%, thisdropped to 8% in 2004.14

My argument of a dual hiring strategy has two centralscope conditions. First, politicians must have a preference togovern a state that functions at least reasonably well. Thispreference encourages them to recruit bureaucrats for high-ranked jobs based on nonpartisan criteria. This preference islikely to be intensified in competitive electoral environmentsbecause the electoral rewards (or punishment) for deliver-ing (not delivering) economic stability or development havegreater consequences; incumbents may be ousted.15 Second,politicians must mobilize electoral support through grass-roots brokers. This condition is likely to hold in developingdemocracies where parties rely on direct personal contact torally voters. As organizationally strong political parties arelikely to rely more on partisan as opposed to communitybrokers (Baldwin 2013; Gottlieb 2017), the pressure to re-ward party brokers with public sector jobs is expected to behigher in countries with organizationally strong politicalparties.

Ghana is a case in which there is evidence of performancevoting and high levels of electoral competition. Furthermore,both of the two major political parties are organizationallystrong and rely on party brokers to mobilize support (Bob-Milliar 2012; Brierley and Nathan 2020; Ichino and Nathan2013). Regarding performance voting, many scholars haveargued that in Ghana, “popular evaluations of governmentperformance trump the pervasive tugs of language and tribe”(Bratton et al. 2012, 30). For example, Lindberg and Morrison(2008, 121) conduct an open-ended survey and differentiatebetween evaluative and nonevaluative motivations for votechoice and conclude that “evaluative voting behavior is by farthe most common stance in Ghana.” Similarly, Jeffries (1998,


205) argues that “the basis for the NDC’s victory [in 1996]was laid by its economic and developmental policies,” mostimportantly the Rawlings government’s provision of develop-ment projects. The influential role of economic performance onvoters’ attitudes and eventual vote choice has also been dem-onstrated through quantitative analyses of both the 2000 and2008 elections (Hoffman and Long 2013; Youde 2005).

Political power in Ghana is decentralized to 216 localgovernments: one local government per district.16 Local gov-ernments are responsible for the development of districts, in-cluding the provision of basic infrastructure and public worksand services. The president appoints a district chief executive(akin to a mayor) to head each local government, in consul-tation with the local branch of the ruling party. Thus, allmayors are members of the ruling party, regardless of thepartisanship of the majority of voters in the district. Each localgovernment has a political and a bureaucratic arm. This ar-ticles focuses on the bureaucratic arm. Once appointed, localbureaucrats—in both high-skilled and low-skilled positions—have security of tenure.

DATA SET OF LOCAL BUREAUCRATSAND MEASUREMENTThe data set includes employee-level information of civilservants working in 199 of Ghana’s 216 local governmentswho were in active employment in 2015.17 Over 40,000 bu-reaucrats are included in the data. I restrict the analysis tobureaucrats hired between 2005 and 2012, which leaves atotal of 17,942 recruits. The mean number of workers perdistrict is 191. In theory, all local government employeesshould be included. While the data are as comprehensive aspossible, some gaps remain.18 To analyze hiring patterns, Iuse information of the date that each bureaucrat was hired towork in the local bureaucrat. This date reflects the first re-cruitment date of each bureaucrat.

All candidates for local government jobs are recruitedthrough a centralized hiring process that requires candidatesto submit a generic application form to the local governmentsecretariat offices in the capital city, Accra.19 Candidates can

16. Local governments are known as district, municipal, or metropolitanassemblies, with classifications based on the population of the district.

17. These data were compiled by Ghana’s Ministry of Local Govern-ment and Rural Development and the Local Government Service inpartnership with consultants hired by the European Union.

18. One omission is that the top bureaucrat in a district (the districtcoordinating director) is often excluded.

19. While local governments can recruit temporary workers and paythem using their own internally generated revenues, they are not able torecruit permanent employees.

submit these applications at any time, but many applicantsapply following mass-hiring advertisements that the gov-ernment places in national newspapers. Interviews with staffat the Local Government Secretariat suggest that mass hiringis the modal type of hiring, as opposed to hiring for indi-vidual positions as they become available.

Candidates applying for positions that I code as high-skilled(planning officers, budget officers, engineers, accountants, andeconomists) are typically recruited using interviews and exams.Candidates recruited to low-skilled positions (sanitation of-ficers, laborers, security guards and drivers) do not sit formalexams. Anecdotal evidence suggests that these candidates areemployed through interviews, as well as upon recommendationfrom local governments to the national secretariat. Recommen-dations from local governments are prone to partisan influencebecause mayors are political appointees and thus copartisanwith the national government. In addition to direct recom-mendations, which likely reveal a candidates partisanship,during the hiring process, politicians and bureaucrats whoengage in recruitment can use informational cues—namesand home town and region—to infer the likely partisanshipof each potential new hire.

Measuring levels of meritocracywith administrative dataI use the bureaucrat data combined with a change in thegoverning party to assess patronage versus meritocratic hiring.If hiring were purely meritocratic, changes in the governingparty should not influence the types of bureaucrats who arehired. Conversely, swings in who gets hired as power changeshands would indicate non-merit-based hiring. An additionaladvantage of using administrative data is that it also permits thedisaggregation of civil servants across different departmentsand different types of positions to investigate which types ofjobs or departments are isolated from interference. This ap-proach thus allows me to assess my theory. One challenge ofusing administrative data is the need to first code the inferredpartisanship of each bureaucratic recruit.

Coding civil servants’ partisanshipThe ideal data set with which to assess the theory would in-clude the partisanship of each bureaucrat at the time of hiring.However, the act of collecting such data would implicate thegovernment in discriminatory practices and is therefore un-available. Instead of relying on a partisanship variable, I use theavailable data to create two proxies of partisanship. First, Icode partisanship using the ethnic group of each bureaucrat.


Second, I use their home region.20 The former approachcomplements recent empirical work on bureaucrats in Africathat also uses individuals’ ethnic group to determine partisan-ship (Hassan 2016). The results that follow are robust to usingeither measure.21 In the main analysis, I focus on the resultsbased on individual ethnicity as this is potentially a more pre-cise categorization of individual partisanship.22

As ethnicity was not a variable in the bureaucrat-leveldata, I coded ethnicity based on the first and last names ofeach worker. Names were first split into name fragments,which were then coded into one of seven ethnic groupings bymultiple research assistants in Ghana.23 After matching namesto ethnic groupings, I use Afrobarometer survey data to linkethnic groups to political parties.24 Based on these data I codebureaucrats who are Akans as pro-NPP.25 I code both Ewe andNorthern ethnic groups as being aligned with the NDC.26

Regarding the second proxy, I identify party strongholdsas home regions where the majority of citizens voted for thesame party across both elections I analyze.27 I code pro-NDCbureaucrats as those whose home region is Northern, Volta,

20. Home regions refer to where bureaucrats say they “come from”;

typically these regions are where individuals have familial roots.21. I replicate the results using home region in the appendix, available

online.22. For example, the NPP is dominant in the Ashanti region, which is

the most populous region in the country. While 75% of the population inthe Ashanti region are Akans (who are aligned with the NPP), 25% belongto other ethnic groups. As this example shows, using home region po-tentially misclassifies individuals.

23. The ethnic groupings are as follows: Akan (non-Fante), Akan,Fante, Ewe, Ga-Dangme, Guan, and Northern. The dictionary that I createbuilds on one made by Noah Nathan.

24. Specifically, I use data from rounds 3–6. I do not use data fromrounds 1 and 2 because they did not ask which party the respondentwould vote for if the election were held tomorrow. I drop respondents whodid not answer this question, including those who refused to answer oranswered “don’t know.”

25. See table E.1.26. Ethnicity does not perfectly align with partisanship; not all Akans

(Ewes/Northerners) support the NPP (NDC). There could be concern thatmeasurement error biases in favor of finding evidence of patronage hiringin instances in which some Ewe and Northern bureaucrats who are codedas being Pro-NDC are in fact supporters of the opposition. However, thereis just as likely to be measurement error in the other direction: some of theAkans hired by the NDC that are coded as NPP supporters will in fact beNDC supporters, leading me to underestimate patronage hiring. Ex ante,there is no reason to expect that bias in one direction will be more sub-stantial than in the other.

27. Table E.2 displays the election results disaggregated by region. TheNDC captured the majority of votes in the Volta, Northern, Upper West,and Upper East regions, while the NPP received the majority of the votesin the Ashanti and Eastern regions.

Upper West or Upper East, and pro-NPP bureaucrats asthose from either the Ashanti or Eastern regions.

EMPIRICAL STRATEGYTo assess evidence of partisan bias in public sector hiring, Iuse the change in Ghana’s ruling party following the De-cember 2008 elections. The NDC’s John Evans Atta-Millswon the 2008 presidential election, and his party won amajority of seats in the parliament. Ghana’s alternation in2008 can be used as a cut point around which to investigatepotential changes in bureaucratic hiring. If bureaucrats arerecruited on the basis of their partisan ties, we would expectto see significant changes in the partisanship of bureaucratswho were hired after the new party came to power. Specifi-cally, if this is the case, we should expect to find a drop in theshare of pro-NPP bureaucrats hired after 2008 and a cor-responding increase in the share of pro-NDC bureaucratshired. My main hypothesis predicts that any increase in pro-NDC hires would be confined to low-skilled positions.

To investigate the relationship between bureaucrats’partisanship and the ruling party, I run the following logisticregression model:

logit(pi) p b0 1 b1Period 2i 1 b2Low�skilledi

1 b3Period 2 # Low�skilledi 1 b4Xi 1 ϵi:

Here pi is a dummy variable that takes the value 1 when thebureaucrat is aligned with the NDC (NPP). Period 2 identifiesthe second hiring period after the change in the ruling partyfollowing the 2008 election. Low-skilledi is a dummy variablethat takes the value 1 when the position is low ranking and 0otherwise. I code positions using information on the job titleof each bureaucrat. All positions are classified as either high-skilled or low-skilled.

A summary list of positions and how they are coded ispresented in table B.1 (tables A.1, B.1–B.5, C.1, E.1, E.2, G.1–G.3, H.1, J.1–J.3, L.1, M.1, and P.1 are available online). Theprocess of classifying the positions is subjective. My aim wasto code low-skilled positions as those that it would be un-controversial to describe as such.28 Tables B.2–B.5 show thejob titles, frequencies, and coding for all positions. The dataalso contain information on the gender, age, and highesteducational level of each bureaucrat. Thus, Xi is a matrix thatcontains these control variables.

28. In app. sec. L, I rerun the analysis varying the classifications of anumber of positions where the classification may be considered contro-versial. The results are robust to these reclassifications. I also conduct theanalysis dropping one position at a time to check that the results are notdriven by recruitment to a single position; again the results remain robust.


Additional information on civil servant databaseWhile the bureaucrat-level data that I analyze are unusuallyrich, one drawback is that they were collected at a singlepoint in time (in June 2015). Therefore, I do not have in-formation about bureaucrats who were hired and vacatedtheir positions before the data were collected.29 One concernis that pro-NPP bureaucrats hired by the NPP governmentresigned when the NDC came to power in 2008. The effect ofthis attrition would be to diminish the likelihood that Iwould find evidence of partisan hiring under NDC rule. Thisis because pro-NPP bureaucrats, some of whom drop out ofthe data, would appear to compose a smaller share of thosehired by the NPP. Correspondingly, pro-NDC bureaucratswould appear to compose a larger share of those hired in thefirst period.

There are three reasons why this type of attrition might beunlikely. First, civil service jobs are valuable to workers be-cause they offer financial stability in the form of a monthlypaycheck, benefits, and an assured pension. Civil servantscan also ask politicians for help when they face emergencies.Therefore it is unlikely that an employee would give up theirjob just because they did not support the new ruling party.Second, while bureaucrats may be unhappy with the changein government, they are likely to anticipate that their pre-ferred party will not be out of office for too long. Indeed,Ghana has seen alternations of power between the two majorparties every eight years since its return to democracy in1992. Third, if it were the case that bureaucrats resigned enmasse following the 2008 elections, the data should show aspike in hiring in 2009 (or 2010) to make up for the deficit ofworkers. Figure D.1 (figs. D.1, F.1, I.1, K.1, N.1, O.1, and Q.1are available online) plots the total number of hires per yearbetween 2005 and 2012 and shows that there were fewerhires in 2009 and 2010 than in 2008. These figures indicatethat there was not a mass departure from the bureaucracyafter the 2008 election.

RESULTSDescriptive analysesTo assess my argument, I first present descriptive data thatdisplay the characteristics of bureaucrats hired across thetwo electoral periods. Second, I conduct a series of logisticregressions that control for a number of important potentialconfounding variables that may influence hiring.

Table 1 shows the share of NDC and NPP bureaucratshired in both periods and presents the results of a difference-in-means test. Bureaucrats’ partisanship is coded based on

29. This is the first database of employees working in local governmentsin Ghana, which makes it impossible to consult older databases.

their ethnicity.30 The results show a positive and statisticallysignificant increase in the share of NDC bureaucrats hired tolow-skilled positions in the second period (4.56 percentagepoints, p p :000). The results also display a statistically sig-nificant reduction in the share of NPP bureaucrats hired forlow-skilled positions (24.83 percentage points). The changein the governing party is not associated with any changes inthe share of NDC or NPP bureaucrats hired to high-skilledpositions (shown by the insignificant p-values). Overall, theaggregate data lend initial support to the main hypothesis—that the NDC government favored its copartisans when se-lecting candidates for low-skilled positions. In addition, thenew government appears to disfavor NPP bureaucrats forlow-skilled positions.

In figure 1, I categorize bureaucrats into three types: pro-NDC, pro-NPP, and swing. The swing category indicatesbureaucrats who do not belong to a politically aligned ethnicgroup.31 Each plot displays the share of hires in each categoryper year. Figure 1A displays trends for high-skilled positions,while figure 1B displays trends for low-skilled positions. Thedotted lines indicate the change in the ruling party in De-cember 2008.

Figure 1A provides evidence that the partisanship of bu-reaucrats recruited to high-skilled positions is fairly staticacross the two electoral periods. Under both governments, thelargest share of new hires was from the main ethnic groupaligned with the NPP—the Akans. This makes sense becausethe Akans are the largest single ethnic group in the country.Consistent with table 1, the change in governing party does notappear to alter hiring patterns for high-skilled jobs.

Table 1. Difference-in-Means Test (Ethnic Group Coding)

30. Table H.1 disregion. The main ressistent in both tables

31. The unaligne

Period 1(% of hires)

plays the sameult of partisan.d ethnic groups

Period 2(% of hires)

test with bureaubias for low-ski

are Fanti, Ga-D

Difference

crats coded by hlled positions is

angme, and Gu

p

NDC bureaucrat:
Low-skilled 35.25 39.81 4.56 .000 High-skilled 33.69 32.29 21.4 .164
NPP bureaucrat:
Low-skilled 44.55 39.72 24.83 .000 High-skilled 48.68 48.25 2.43 .687
Note. NDC p National Democratic Congress; NPP p New PatrioticParty.

omecon-

an.

33. The ideal way to investigate meritocracy is to collect individual-

Figure 1B presents the trend for low-skilled positions.In contrast to high-skilled jobs, there was a steady increase inthe share of NDC bureaucrats hired for low-skilled positionsafter the NDC government was elected at the end of 2008. Thesize of the increase was 10 percentage points over the period,from 36.5% at the start of their term to 46.5% by the end of theterm. The data also suggest that much of this increase tookplace at the direct expense of NPP candidates, who experi-enced a corresponding decline in hires.32 When the NPP werein office they also appear to have increasingly hired Pro-NPPand swing bureaucrats to low-skilled positions in the run up tothe December 2008 election. This was potentially a strategy tosecure their victory or a result of the expectation that they maysoon be out of office. In short, the yearly trends complementthe aggregate results in table 1 and suggest partisan hiring forlow-skilled positions.

Evidence of meritocratic hiring for high-skilled posi-

tions. The relatively static pattern of hiring among differenttypes of bureaucrats for professional positions across variousgoverning parties suggests that nonpartisan criteria drive re-

32. Fig. I.1 displays the same plot with bureaucrats coded according totheir home region. The results remain the same, with a sharp increase inbureaucrats with traits that aligned them with the NDC supporters hiredafter the change in governing party.

cruitment to high-skilled positions. Two additional pieces ofevidence lend support to the claim that bureaucrats are hiredon the basis of their merits and suitability for these high-ranked positions.33

First, recruits to high-skilled jobs become increasinglywell qualified: the share of new hires who hold bachelor’sdegrees increases nearly every year from about 8% in 1975 to60% in 2012 (see fig. F.1). Restricting this analysis to the twohiring periods in question, 35% of people recruited to high-skilled jobs by the NPP government (2004–8) held either abachelor’s or master’s degree, compared to 55% for NDC gov-ernment hires (2008–12). This suggests that as access to ter-tiary education expanded in Ghana, the government recruitedincreasingly well-qualified applicants.

Second, recruitment through competitive procedures alsobecame increasingly common. A separate survey that I con-ducted with 864 local bureaucrats in high-skilled positionsrevealed that the vast majority were selected through

Figure 1. Share of bureaucrat types across two electoral periods (2005–8, 2009–12). Employees are categorized as pro-NDC, pro-NPP, or swing. Each plot

displays the share of new hires from each group. Bureaucrats are coded according to their ethnic group. A, Trends for high-skilled positions; B, trends for

low-skilled positions. The dotted line corresponds to the election of the NDC in December 2008. NDC p National Democratic Congress; NPP p New

Patriotic Party.


level data on the pool of applicants, as well as data on who is eventuallyhired. I was not able to obtain data on the pool of applicants across thisperiod of time. To my knowledge, there are no studies of African bu-reaucracies that perform such an analysis. An important exception in thecontext of Latin America is Dal Bó, Finan, and Rossi (2013), who doobtain data on the applicant pool.


competitive processes (Brierley 2020).34 Restricting the analy-sis to the two hiring periods, 33% of bureaucrats hired by theNPP government took a competitive exam, while 92% had aninterview. This compares to 67% who took a competitiveexam and 96% who had an interview under the NDC govern-ment.35 Overall, while the results do not point conclusively tomeritocracy, they suggest that when the NDC governmentcame to office after the 2008 election, it used increasinglymeritocratic hiring criteria to select highly educated indi-viduals through increasingly competitive procedures.

Regression analysesA series of regression analyses adds further credibility to theresults presented above (table 2). In these models, the de-pendent variables are dummy variables that indicate bureau-crats’ partisan type. Column 1 predicts pro-NDC bureaucrats,column 2 predicts swing bureaucrats, and column 3 predictspro-NPP bureaucrats. The two main explanatory variables area dummy variable that distinguishes between the two hiringperiods—the change in ruling party—and an indicator of

34. These bureaucrats worked in a random sample of eighty localgovernments across the country. This survey took place in 2015 and 2016.

35. I acknowledge that the conduct of interviews and exams does notensure meritocracy. I do not have data on how candidates performed inthese assessments. Furthermore, candidates’ performance, especially ininterviews, is somewhat subjective. However, the fact that the NDC gov-ernment hired more candidates following interviews and exams suggeststwo things. First, it suggests a move toward meritocracy. Second, thedelegation of hiring processes from politicians to bureaucrats: bureaucratsconduct these processes and evaluate candidates.

whether the job is low-skilled. A positive coefficient onthe interaction term would indicate that the new ruling partydistributed more low-skilled than high-skilled posts to theircopartisans. In these models, I hold constant gender, age, andhighest level of education.36

In table 2 column 1, the coefficient on the variable thatindicates a change in the ruling party is negative. This sug-gests that a change in the ruling party is associated with anoverall decrease in the likelihood of a pro-NDC bureaucratbeing hired to a high-skilled position. The next coefficientindicates the relationship between low-skilled jobs and beinga pro-NDC bureaucrat; it is also negative. The coefficient onthe interaction term is the key quantity of interest. The pos-itive sign on this term shows that the change in government isassociated with an increase in the likelihood of a pro-NDCbureaucrat being hired for a low-skilled position. This coef-ficient is statistically significant below the 1% level. Figure 2illustrates the substantive significance of this result.

Table 2 column 2 displays the results of the same modelspecification with swing bureaucrats as the dependent vari-able. None of the coefficients are statistically significant,which suggests that the change in the ruling party did notaffect hiring patterns for politically nonaligned bureaucrats.

Finally, table 2 column 3 displays the results of the samemodel specification with pro-NPP bureaucrats as the dependent

Table 2. Logistic Regression Predicting Hiring of Partisan Bureaucrats across Each Time Period

NDC Bureaucrat(1)

36. MaIn the appeage, and thvations inc

Swing Bureaucrat(2)

ny bureaucrats did not indicndix, I present the same resue results remain the same.reases by roughly 8,000 bure

NPP Bureaucrat(3)

Change in ruling party
2.154*** .119 .083 (.059) (.092) (.056)
Low-skilled
2.129 .202 .010 (.094) (.135) (.084)
Change in ruling party#low-skilled
.614*** 2.187 2.434*** (.103) (.150) (.094)
Observations
9,780 9,068 9,780 Log likelihood 26,111.116 23,284.289 26,725.204
Note. Regressions control for gender, age at time of hiring, and highest level of education. Bureaucrats’ partisanship is codedaccording to their ethnicity. Change in the ruling party follows Ghana’s December 2008 election, which resulted in the NDCcoming to office. Standard errors in parentheses. NDC p National Democratic Congress; NPP p New Patriotic Party.* p ! .1.** p ! .05.*** p ! .01.

ate their highest level of education.lts controlling only for gender andIn table J.1 the number of obser-aucrats.

37. The total number of districts went from 138 to 170 (see Ayee 2013).


vcnpthsths

sblod9

nwth

ariable. The first coefficient is positive, although not statisti-ally significant. This suggests that the NDC government didot demonstrate partisan bias when recruiting to high-skilledositions; they continued to hire pro-NPP bureaucrats intoese positions. However, there is a negative and statistically

ignificant coefficient on the interaction term, which suggestsat NPP bureaucrats were less likely to be hired for low-

killed positions.Figure 2 displays the substantive significance of the re-

ults. I calculate the predicted probabilities of a pro-NDCureaucrat being hired to (a) a high-skilled position and (b) aw-skilled position in the two time periods. Figure 2 shows theifference in these predicted probabilities and the associated5% confidence intervals.

Figure 2A demonstrates that the change in government isot associated with a change in the likelihood of pro-NDCorkers being hired for high-skilled positions. In other words,ere is no evidence of discriminatory partisan hiring for high-

skilled jobs. In contrast, pro-NDC bureaucrats were 10 per-centage points more likely to be hired for low-skilled posi-tions after the NDC came to power at the end of 2008. Theseresults support the first hypothesis and demonstrate that theregression results are substantively important.

Figure 2B suggests that when the NDC was in power, itwas less likely to hire pro-NPP bureaucrats to low-skilledpositions. The change in predicted probabilities is about 8 per-centage points. The NDC government was neither more norless likely to hire pro-NPP bureaucrats for high-skilled po-sitions. In summary, the results suggest that the partisan-ship of bureaucrats did not influence who got hired for high-skilled positions, but did influence selection into low-skilledpositions.

An alternative argument to the one that I present is that,besides the change in ruling party, something else occurredthat affected hiring patterns. Experts on Ghanaian politics,for example, might point to the creation of 32 new dis-tricts and corresponding local governments, in February 2008(10 months before the election).37 If these new districts wereconcentrated in areas where the NDC were electorally dominant,one could argue that the new ruling party were disproportion-ately hiring pro-NDC bureaucrats into low-skilled positionsto staff these new local offices. However, this argument is anunlikely explanation of the results because the new districtswere not concentrated in regions that I code as being Pro-NDC. Indeed, only 10 of the new local governments (28%)were in the three northern regions or in the Volta region.

A similar change occurred in 2012, when the NDC gov-ernment created another 46 districts. In this case, a largershare of the new local governments were in the regions that Icode as being Pro-NDC (44%). As a robustness check, I rerunthe analysis dropping all hires made after June 28, 2012 (thedate the new districts were created). The results are robust tothis specification (see table J.3).

Another rival argument is that the patterns I find resultfrom an under supply of available educated citizens fromethnic groups that are aligned with the NDC government.This would imply that the NDC government had no choicebut to hire educated citizens from either nonaligned ethnicgroups or from groups aligned to the NPP. Table G.1 dis-plays evidence that there was a roughly equal available sup-ply of highly educated citizens aligned with either party. Spe-cifically, I estimate that there were a total of approximately247,470 NDC-aligned citizens with higher levels of educationcompared to 271,700 NPP-aligned citizens. The NDC gov-ernment recruited a total of 5,940 professionals into the local

Figure 2. Difference in the predicted probability of a pro-NDC and pro-NPP

bureaucrat being hired in each term (2005–8 and 2009–12), disaggregated

by job type into low-skilled and high-skilled positions. These predicted

probabilities are calculated using the coefficients in table 2, columns 1 and

2. NDC p National Democratic Congress; NPP p New Patriotic Party.


bureaucrats during their first term in office after the 2008election (see table C.1). Thus, the data suggest that had theNDC wanted to hire copartisans to these positions there was alarge enough supply of competent individuals.38

Evidence that low-skilled positions are givento party intermediariesThe above results suggest that the new NDC governmentrecruited copartisans into low-skilled positions after theycame to office in 2008. In this section, I present quantitativeand qualitative evidence that politicians in Ghana rewardparty brokers with low-skilled positions in the public sector,including in local governments.

What motivates party brokers to work for the party

machine? Citizens in Ghana believe that brokers work forparties in return for material benefits. A country-specificquestion in the 2012 Afrobarometer asked citizens why theythought party intermediaries work for parties. Roughly 60% ofGhanaians responded that they do so because “they expectmaterial rewards after winning political power” (Armah-Attoh 2017, 2).39 The brokers themselves admit that they workfor parties because of the selective incentives they offer, in-cluding jobs (Bob-Milliar 2012). Bob-Milliar (2012, 677) ex-plains, “throughout the Fourth Republic, party foot-soldiershave demanded openly to be rewarded by their respectiveparties for their activism.” Pertinent to this article is whetherlow-skilled jobs in local governments are rewarded to partyintermediaries.

I subset the data to analyze the share of pro-NDC hires tovarious low-skilled positions between 2008 and 2012. Posi-tions that had large shares of copartisans include environ-mental assistants (65% of all hires), watch guards (55%), anddisaster management officers (39%).40 These findings are con-sistent with accounts from Ghanaian scholars and journalistswho have also noted, for example, that local disaster manage-ment jobs are often distributed to party intermediaries. One re-porter explains that disaster management “has been known to bea den for political appointees with governments recruiting its sup-porters to the organization even though they have little or noexperience in disaster management” (Peace FM Online 2017).

38. A final rival argument is that the patterns that I find are the resultof changes in the supply of recruits, rather than a change in demand forcertain types of workers. However, given the immense value of having apublic sector job, and the fact that individuals do not have to pay to apply,I do not expect that pro-NPP recruits would stop applying after thechange in government.

39. In comparison, about one-third (31%), believe brokers toil fortheir party because “they believe in their programs.”

40. See table M.1.

In summary, the results are consistent with scholarly and jour-nalistic accounts, as well as with public opinion, that parties inGhana reward intermediaries with low-skilled public sector jobs.

What types of districts get low-skilled hires? The liter-ature on patronage suggests that parties distribute publicsector jobs to loyal supporters as rewards for their work ingetting out the votes (Robinson and Verdier 2013). If poli-ticians do distribute jobs as rewards we might expect to seemore hires to low-skilled positions in districts where theparty experienced significant increases in their vote sharecompared to the prior election. Such a result would indicatethat the incumbent party is strategic in their distribution ofappointments; rewarding brokers who brought them votes.

The results support this interpretation of low-skilledjobs being distributed as rewards. Table 3 displays a positivecorrelation between the number of hires to low-skilled jobsand district-level changes in the vote share of the NDC. Thedata are aggregated at the district level, such that the depen-dent variable is the number of hires to low-skilled positionsper district. The data are restricted to the four-year electoralterm after the NDC came to power following the December2008 election in order to analyze hires made by the NDCgovernment.41 The main independent variable is the change inthe vote share of the incumbent party between the 2004 and2008 elections.

Columns 1–3 in table 3 consider all hires to low-skilledjobs. Columns 4–6 consider hires who belong to ethnicgroups aligned with the ruling party. All regressions includeregion fixed effects. Column 1 shows a positive and statis-tically significant association between change in the rulingparty vote share and the number of hires to low-skilledpositions. Column 4 also shows a positive association be-tween change in ruling party vote share and the number ofpro-NDC hires, although this coefficient is not statisticallysignificant.42 Columns 2 and 4 control for both the type oflocal government and the population of the district.43 In-cluding these variables significantly improves the fit of themodel, and these variables are highly predictive: municipaland metropolitan districts have more hires.44 In columns 5and 6 the coefficient on change in NDC votes share also

41. There is significant variation in the number of hires per district,ranging from 1 to 613 recruits, with a mean of 29 and a median of 22 perdistrict.

42. The p-value on this coefficient is .17.43. Local governments in Ghana are categorized as district, municipal,

or metropolitan assemblies based on the population of the district.44. The negative coefficient on the population coefficient is most

likely because the type of local government already accounts for thevariation in population across districts.


becomes statistically significant (at the .1 level). Columns 3and 6 also include the share of houses with electricity, whichserves as a proxy for urbanization. Again, the coefficient ofinterest remains significant and positive. Substantively, theresults suggest that a district where the party saw a 10 per-centage point increase in vote share would get five morehires than a district that saw a 10 percentage point decreasein ruling party vote share (see fig. Q.1 for a plot of the marginaleffect of vote share). Overall, this analysis points to the strategicallocation of low-skilled positions to reward party interme-diaries who rallied support for the new incumbent party.45

Is there evidence that recruits to low-skilled positions

are embedded in the communities they are hired to work

in? Finally, I focus on one characteristic of party brokersthat is essential for them to perform their roles—their embed-dedness in the community. I use the distance that each officerworks from her hometown as a proxy for entrenchment. I assesswhether bureaucrats in low-skilled positions work in localgovernments closer to their home towns than bureaucrats in

45. Table P.1 presents the same analyses with Pro-NPP hires as thedependent variable. These results show that there is no relationship be-tween the change in NDC vote share and the number of Pro-NPP hires.

high-skilled positions. To clarify, each district has one set oflocal government offices where all bureaucrats work. These of-fices are located in the district capitals. Thus, workplace loca-tions are constant across both groups of bureaucrats. Workingclose to one’s hometown is not sufficient evidence that a bu-reaucrat in a low-skilled position is a broker. However, it is likelyto be a necessary condition, and thus evidence of a differenceacross job type lends further credibility to my argument.

I use the hometown variable in the data set to identify thetown that bureaucrats view as their “home,” which typicallyrefers to the place where they have familial roots. I geocodethe locations of district capitals and home towns and cal-culate the distance between the two for each bureaucrat inthe data.46 The results show that the median distance be-tween home town and work location is significantly lower forbureaucrats in low-skilled compared to high-skilled posi-tions.47 The median distance from their hometown is 36 kilo-meters for low-skilled and 92 kilometers for high-skilled

Table 3. Predictors of the Number of Hires to Low-Skilled Positions (2009–12)

Total Hires

46. I geocusing the ggmin Ghana. I ocapital) for 74bureaucrats).

47. See fig

Pro-NDC Hires

(1)
(2) (3) (4)
ode locations uap package inbtain complet% of the burea

. N.1.

(5)

sing search requeR. Google Maps de matches (i.e., bucrats hired in the

(6)

D in NDC vote share
36.146** 24.995** 22.492** 18.471 10.453* 12.215* (16.660) (10.595) (10.807) (13.391) (6.080) (6.180)
Metropolitan
168.005*** 165.177*** 161.024*** 163.029*** (12.578) (12.805) (7.136) (7.246)
Municipal
16.318*** 15.576*** 8.473*** 8.978*** (2.359) (2.444) (1.345) (1.386)
log(Population)
23.328 24.164* 21.526 2.901 (2.038) (2.163) (1.169) (1.244)
Electricity share
8.799 26.326 (7.688) (4.417)
Region fixed effects
Yes Yes Yes Yes Yes Yes Observations 157 157 157 152 152 152 R2 .151 .667 .670 .254 .851 .853 Adjusted R2 .093 .637 .638 .201 .837 .838
Note. Unit of analysis is the district. Change in National Democratic Congress (NDC) vote share is measured between the2004 and 2008 elections using constituency-level data from the parliamentary elections. Control variables (population andelectricity share) are from Ghana’s 2010 Housing and Population Census. Standard errors in parentheses.* p ! .1.** p ! .05.*** p ! .01.

sts to the Google Maps APIoes not include every town

oth hometown and districtstudy period (N p 13; 283


jobs.48 Figure O.1 focuses on a few select positions to furtherdemonstrate these differences. The results show that budgetofficers, planning officers, and assistant directors work, onaverage, much farther from their home communities thandrivers, laborers, and disaster management assistants. Themedian distance from hometown is 19 kilometers for labor-ers, 21 kilometers for drivers, and 28 kilometers for disasterassistants, compared to 151 kilometers for budget officers,161 kilometers for planning officers, and 174 kilometers forassistant directors.

These results lend further credence to my theory in twoways. First, they show that bureaucrats working in high-skilled positions are unlikely to be party brokers, suggestinginstead that they are hired based on nonpartisan criteria. Sec-ond, and conversely, the results provide evidence that bureau-crats in low-skilled positions could work as party inter-mediaries. Working close to one’s hometown allows partisanrecruits to mobilize support for the incumbent by utilizingtheir social ties with local communities and local party elites.Brokers employed in the district government may be par-ticularly effective intermediaries, as they can help citizensaccess the state and can distribute state resources to localvoting blocs (Auerbach and Thachil 2018).

CONCLUSIONThe improvement of bureaucratic capacity is essential toenhancing state capacity in developing democracies. Accord-ingly, who the state hires into the public sector has broadimplications for development. Many governments spend thebulk of public expenditure on public sector salaries. This ar-ticle assesses meritocratic versus patronage hiring in the con-text of Ghana. I theorize that the costs to politicians of com-petitive hiring are not constant across positions. Conceptualizingmeritocracy as a continuous variable, I do not ask whetherjobs are distributed on the basis of merit but which jobs areawarded competitively. I theorize that politicians in clien-telistic democracies will be more willing to relinquish controlover hiring for high-skilled jobs than they will be for low-skilled positions in the public sector. Politicians reward partybrokers who mobilize on their behalf with these low-skilled jobs.

I analyze an original individual-level data set on the uni-verse of bureaucrats working across about 200 local govern-ments in Ghana. As party affiliation is not a variable in thedata set, I code bureaucrats’ partisanship based on their eth-nicity and home region. I use a change in the ruling partyfollowing Ghana’s 2008 elections to compare patterns of hir-ing under two opposing parties. The results suggest that pol-

48. Given the skewed distribution, the median is the most appropriatemeasure of average trends.

iticians are more likely to interfere in hiring for low-skilledjobs than high-skilled jobs: there is a 10 percentage point in-crease in the probability of a copartisan being hired to a low-skilled position after the change in government.

My theory highlights the problems of using legislation tomeasure meritocracy (Geddes 1994; Ting et al. 2013). Meritlegislation has been shown to be neither necessary nor suf-ficient to ensure nonpartisan hiring recruitment (Schuster2017). Indeed, weak rule-of-law and legal loopholes meanthat politicians regularly make partisan appointments inspite of merit laws (Meyer-Sahling 2006). The results suggestthe need for continuous measures of patronage and meri-tocracy. As an alternative to a legislative analysis, I proposemeasuring levels of patronage using administrative data onthe composition of public workforces.

The findings also shed light on the mixed results in theliterature on the effect of competition on patronage hiring.Driscoll (2017) finds that electorally competitive districtshave more local government employees in Ghana. He sim-ilarly argues that politicians exchange these—mainly low-skilled—positions with party activists. Pierskalla and Sacks(2019) show that the introduction of elections increased thenumber of teachers employed in Indonesia. In contrast, muchof the seminal literature on the state suggests that electoralcompetition can constrain runaway state building and in-centivize politicians to support meritocracy (Geddes 1994;O’Dwyer 2006). My results suggest that electoral competitioncan have both of these effects.

This study has significant implications for future researchon the public sector. As new data become available, it wouldbe fruitful to replicate the analysis in other contexts. Similarbureaucrat-level data can be used to assess meritocracy inother countries, and across different public sector departments.Importantly, my results suggest the need to distinguish betweendifferent types of jobs in such analyses in order to assess morenuanced hypotheses regarding the effects of democracy onmeritocratic versus patronage hiring practices.

ACKNOWLEDGMENTSFor helpful comments, I thank Brian Crisp, Stefano Fiorin,Barbara Geddes, Mai Hassan, Martha Johnson, NoahNathan, George Ofosu, Jan Pierskalla, Dan Posner, MargitTavits, Mike Thies, Daniel Triesman, Andrea Vilán, andMartha Wilfahrt. I also thank three anonymous reviewers fortheir feedback. Previous versions of this manuscript werepresented at the African Studies Association Annual Con-ference in November 2017, the Data Driven Developmentfor Africa workshop at Indiana University in May 2018, andthe American Political Science Association Annual Confer-ence in November 2018. For sharing their data, I thank the


Local Government Service Secretariat in Accra. I am alsograteful to Gangyi Sun for excellent research assistance.

REFERENCESAfrica Confidential. 2018. “Tax and Spend Dilemmas.” https://www.africa

-confidential.com/article/id/12322/Tax_and_spend_dilemmas.Aljazeera. 2014. “Kenya’s Wage Bill Dilemma.” https://www.aljazeera.com

/blogs/africa/2014/03/98646.html.Armah-Attoh, Daniel. 2017. “Political Party Vigilantism: A Growing Fox

Worth Killing before It Mutates into a Monster.” Elections Policy Briefno. 4, Ghana Center for Democratic Development, Accra.

Auerbach, Adam, and Tariq Thachil. 2018. “How Clients Select Brokers:Competition and Choice in India’s Slums.” American Political ScienceAssociation 112 (4): 1–17.

Ayee, Joseph R. A. 2013. “The Political Economy of the Creation of Districts

in Ghana.” Journal of Asian and African Studies 48 (5): 623–45.Baldwin, Kate. 2013. “Why Vote with the Chief ? Political Connections and

Public Goods Provision in Zambia.” American Journal of Political Sci-ence 57 (4): 794–809.

Bayart, Jean-Francois. 1993. The State in Africa: The Politics of the Belly.London: Longman.

Bob-Milliar, George M. 2012. “Political Party Activism in Ghana: FactorsInfluencing the Decision of the Politically Active to Join a PoliticalParty.” Democratization 19 (4): 668–89.

Bratton, Michael, Ravi Bhavnani, and Tse-Hsin Chen. 2012. “Voting In-tentions in Africa: Ethnic, Economic or Partisan?” Commonwealth andComparative Politics 50 (1): 27–52.

Brierley, Sarah. 2020. “Unprincipled Principals: Co-opted Bureaucrats andCorruption in Ghana.” American Journal of Political Science 64 (2): 209–22.

Brierley, Sarah, and Noah L. Nathan. 2020. "The Connections of Party Brokers."Journal of Politics (forthcoming).

BusinessTech. 2018. “Government’s Insane Wage Bill.” https://businesstech.co.za/news/government/235879/governments-insane-wage-bill/.

Callen, Michael, Saad Gulzar, and Arman Rezaee. 2020. “Can PoliticalAlignment Be Costly?” Journal of Politics 82 (2): 612–26.

Calvo, Ernesto, and Maria Victoria Murillo. 2013. “When Parties MeetVoters: Assessing Political Linkages through Partisan Networks andDistributive Expectations in Argentina and Chile.” Comparative Polit-ical Studies 46 (7): 851–82.

Carlson, Elizabeth. 2015. “Ethnic Voting and Accountability in Africa: AChoice Experiment in Uganda.” World Politics 67 (2): 353–85.

Carpenter, Daniel P. 2001. The Forging of Bureaucratic Autonomy: Rep-utations, Networks, and Policy Innovation in Executive Agencies, 1862–1928. Princeton, NJ: Princeton University Press.

Chabal, Patrick, and Jean-Pascal Daloz. 1999. Africa Works: The PoliticalInstrumentalization of Disorder. Bloomington: Indiana University Press.

Clements, Benedict J., Sanjeev Gupta, Izabella Karpowicz, and ShamsuddinTareq. 2010. Evaluating Government Employment and Compensation,

vol. 10. Washington, DC: International Monetary Fund.Colonnelli, Emanuele, Mounu Prem, and Edoardo Teso. 2018. “Patronage

and Selection in Public Sector Organizations.” SSRN. https://papers.ssrn.com/soL3/papers.cfm?abstract_idp2942495.

Dal Bó, Ernesto, Frederico Finan, and Martín A. Rossi. 2013. “Strength-ening State Capabilities: The Role of Financial Incentives in the Call toPublic Service.” Quarterly Journal of Economics 128 (3): 1169–218.

Driscoll, Barry. 2017. “Why Political Competition Can Increase Patron-age.” Studies in Comparative International Development 53:404–27.

Ferree, Karen E. 2006. “Explaining South Africa’s Racial Census.” Journalof Politics 68 (4): 803–15.

Fisman, Raymond, and Miriam A. Golden. 2017. Corruption: What Ev-eryone Needs to Know. Oxford: Oxford University Press.

Geddes, Barbara. 1991. “A Game Theoretic Model of Reform in LatinAmerican Democracies.”American Political Science Review 85 (2): 371–92.

Geddes, Barbara. 1994. Politician’s Dilemma: Building State Capacity inLatin America. Berkeley: University of California Press.

Golden, Miriam Brian Min. 2013. “Distributive Politics around theWorld.” Annual Review of Political Science 16:73–99.

Gottlieb, Jessica. 2017. “Explaining Variation in Broker Strategies: A Lab-in-the-Field Experiment in Senegal.” Comparative Political Studies 50(11): 1556–92.

Gottlieb, Jessica, Guy Grossman, Horacio A. Larreguy, and Benjamin Marx.2018. “A Signaling Theory of Distributive Policy Choice: Evidence fromSenegal.” Jounal of Politics 81 (2): 631–47.

Green, Elliott. 2010. “Patronage, District Creation, and Reform in Uganda.”Studies in Comparative International Development 45 (1): 83–103.

Grindle, Merilee S. 2012. Jobs for the Boys: Patronage and the State inComparative Perspective. Cambridge, MA: Harvard University Press.

Grzymala-Busse, Anna. 2007. Rebuilding Leviathan: Party Competitionand State Exploitation in Post-Communist Democracies. Cambridge:Cambridge University Press.

Harding, Robin. 2015. “Attribution and Accountability: Voting for Roadsin Ghana.” World Politics 67 (4): 656–89.

Hassan, Mai. 2016. “The Strategic Shuffle: Ethnic Geography, the InternalSecurity Apparatus, and Elections in Kenya.” American Journal ofPolitical Science 61 (2): 382–95.

Hassan, Mai, and Ryan Sheely. 2017. “Executive-Legislative Relations,Party Defections, and Lower Level Administrative Unit Proliferation:

Evidence from Kenya.” Comparative Political Studies 50 (12): 1595–631.Hoffman, Barak D., and James D. Long. 2013. “Parties, Ethnicity, and

Voting in African Elections.” Comparative Politics 45 (2): 127–46.Ichino, Nahomi, and Noah L. Nathan. 2013. “Do Primaries Improve Electoral

Performance? Clientelism and Intra-Party Conflict in Ghana.” AmericanJournal of Political Science 57 (2): 428–41.

Iyer, Lakshmi, and Anandi Mani. 2012. “Traveling Agents: Political Changeand Bureaucratic Turnover in India.” Review of Economics and Statistics94 (3): 723–39.

Jeffries, Richard. 1998. “The Ghanaian Elections of 1996: Towards theConsolidation of Democracy?” African Affairs 97 (387): 189–208.

Johnson, Ronald N., and Gary D. Libecap. 1994.TheFederalCivil ServiceSystemand the Problem of Bureaucracy. Chicago: University of Chicago Press.

Kopeckỳ, Petr. 2011. “Political Competition and Party Patronage: Public Ap-pointments in Ghana and South Africa.” Political Studies 59 (3): 713–32.

Larreguy, Horacio, John Marshall, and Pablo Querubin. 2016. “Parties,Brokers, and Voter Mobilization: How Turnout Buying Depends uponthe Party’s Capacity to Monitor Brokers.” American Political ScienceReview 110 (1): 160–79.

Lewis-Beck, Michael S., and María Celeste Ratto. 2013. “Economic Votingin Latin America: A General Model.” Electoral Studies 32 (3): 489–93.

Lindberg, Staffan I. 2003. “‘It’s Our Time to “Chop”’: Do Elections in AfricaFeed Neo-Patrimonialism Rather than Counter-act It?” Democrati-zation 10 (2): 121–40.

Lindberg, Staffan I., and Minion K. C. Morrison. 2008. “Are AfricanVoters Really Ethnic or Clientelistic? Survey Evidence from Ghana.”Political Science Quarterly 123 (1): 95–122.

Mattes, Robert, and Jessica Piombo. 2001. “Opposition Parties and theVoters in South Africa’s General Election of 1999.” Democratization8 (3): 101–28.

Meyer-Sahling, Jan-Hinrik. 2006. “The Institutionalization of Political Dis-cretion in Post-Communist Civil Service Systems: The Case of Hungary.”Public Administration 84 (3): 693–715.


O’Dwyer, Conor. 2006. Runaway State-Building: Patronage Politics andDemocratic Development. Baltimore: Johns Hopkins University Press.

Ofosu, George Kwaku. 2019. “Do Fairer Elections Increase the Respon-siveness of Politicians?” American Political Science Review 113 (4): 963–79.

Oliveros, Virginia. 2016. “Making It Personal: Clientelism, Favors, and thePersonalization of Public Administration in Argentina.” ComparativePolitics 48 (3): 373–91.

Oliveros, Virginia, and Christian Schuster. 2018. “Merit, Tenure, andBureaucratic Behavior: Evidence from a Conjoint Experiment in theDominican Republic.” Comparative Political Studies 51 (6): 759–92.

Peace FM Online. 2017. “Over 80% of NADMO Staff Have No Training—Outgoing NADMOP Boss.” https://www.peacefmonline.com/pages/local/social/201703/309213.php.

Pierskalla, Jan H., and Audrey Sacks. 2019. “Personnel Politics: Elections,Clientelistic Competition and Teacher Hiring in Indonesia.” BritishJournal of Political Science. https://doi:10.1017/S0007123418000601.

Posner, Daniel N., and David J. Simon. 2002. “Economic Conditions andIncumbent Support in Africa’s New Democracies: Evidence fromZambia.” Comparative Political Studies 35 (3): 313–36.

Rauch, James E., and Peter B. Evans. 2000. “Bureaucratic Structure andBureaucratic Performance in Less Developed Countries.” Journal ofPublic Economics 75 (1): 49–71.

Roberts, Andrew. 2008. “Hyperaccountability: Economic Voting in Cen-tral and Eastern Europe.” Electoral Studies 27 (3): 533–46.

Roberts, Kenneth M., and Erik Wibbels. 1999. “Party Systems and ElectoralVolatility in Latin America: A Test of Economic, Institutional, andStructural Explanations.” American Political Science Review 93 (3): 575–

90.

Robinson, James A., and Thierry Verdier. 2013. “The Political Economy ofClientelism.” Scandinavian Journal of Economics 115 (2): 260–91.

Schuster, Christian. 2017. “Legal Reform Need Not Come First: Merit-Based Civil Service Management in Law and Practice.” Public Admin-istration 95 (3): 571–88.

Sigman, Rachel. 2015. “Which Jobs for Which Boys? Patronage for Po-litical Finance.” Working paper.

Skowronek, Stephen. 1982. Building a New American State: The Expansionof National Administrative Capacities, 1877–1920. Cambridge: Cam-bridge University Press.

Stokes, Susan C., Thad Dunning, Marcelo Nazareno, and Valeria Brusco.2013. Brokers, Voters, and Clientelism: The Puzzle of Distributive Poli-tics. Cambridge: Cambridge University Press.

Ting, Michael M., James M. Snyder, Shigeo Hirano, and Olle Folke. 2013.“Elections and Reform: The Adoption of Civil Service Systems in theUS States.” Journal of Theoretical Politics 25 (3): 363–87.

Van de Walle, Nicolas. 2001. African Economies and the Politics of Per-manent Crisis, 1979–1999. Cambridge: Cambridge University Press.

Wade, Robert. 1982. “The System of Administrative and Political Cor-ruption: Canal Irrigation in South India.” Journal of DevelopmentStudies 18 (3): 287–328.

Weghorst, Keith R., and Staffan I. Lindberg. 2013. “What Drives the SwingVoter in Africa?” American Journal of Political Science 57 (3): 717–34.

Wilson, James Q. 1961. “The Economy of Patronage.” Journal of PoliticalEconomy 69 (4): 369–80.

Xu, Guo. 2018. “The Costs of Patronage: Evidence from the British Em-pire.” American Economic Review 108 (11): 3170–98.

Youde, Jeremy. 2005. “Economics and Government Popularity in Ghana.”

Electoral Studies 24 (1): 1–16.

combining patronage and merit in public sector recruitment

Documents